Classifying and Characterising Multiple Sclerosis Disease Phenotypes with Functional Connectivity and Machine Learning - European Medical Journal

Continue

Classifying and Characterising Multiple Sclerosis Disease Phenotypes with Functional Connectivity and Machine Learning

2 Mins
Neurology
EMJ Neurology 9.1 2021 Feature Image
Authors:
*Maria A. Rocca,1,2,3,4 Paola Valsasina,1 Olga Marchesi,1 Paolo Preziosa,2,3 Diego Sona,5,6 Jacopo Tessadori,5 Muhammad A. Yamin,5,7 Massimo Filippi2,4,8

1. Neuroimaging Research Unit, Division of Neuroscience, IRCCS San Raffaele Scientific Institute, Milan, Italy
2. Neurology Unit, IRCCS San Raffaele Scientific Institute, Milan, Italy
3. Neuroimaging of CNS WM Unit, Division of Neuroscience, IRCCS San Raffaele Scientific Institute, Milan, Italy
4. Vita-Salute San Raffaele University, Milan, Italy
5. Pattern Analysis and Computer Vision, Istituto Italiano di Tecnologia, Genova, Italy
6. Neuroinformatics Laboratory, Fondazione Bruno Kessler, Trento, Italy
7. Department of Electrical, Electronics and Telecommunication Engineering and Naval Architecture, Università degli Studi di Genova, Genova, Italy
8. Neurophysiology Unit, IRCCS San Raffaele Scientific Institute, Milan, Italy
*Correspondence to [email protected]

Disclosure:

Rocca received speaker honoraria from Bayer, Biogen, Bristol Myers Squibb, Celgene, Genzyme, Merck Serono, Novartis, Roche, and Teva; and receives research support from the MS Society of Canada and Fondazione Italiana Sclerosi Multipla. Preziosa received speaker honoraria from Biogen Idec, Novartis, Bristol Myers Squibb, Genzyme, and Excemed; was supported by a senior research fellowship (FISM2019/BS/009); and financed or co-financed with the ‘5 per mille’ public funding. Filippi is Editor-in-Chief of the Journal of Neurology, Associate Editor of Human Brain Mapping, Associate Editor of Radiology, and Associate Editor of Neurological Sciences; has received compensation for consulting services and/or speaking activities from Alexion, Almirall, Bayer, Biogen, Celgene, Eli Lilly, Genzyme, Merck-Serono, Novartis, Roche, Sanofi, Takeda, and Teva Pharmaceutical Industries; receives research support from Biogen Idec, Merck-Serono, Novartis, Roche, Teva Pharmaceutical Industries, Italian Ministry of Health, Fondazione Italiana Sclerosi Multipla, and ARiSLA (Fondazione Italiana di Ricerca per la SLA); was partially supported by a grant (FISM2019/BS/009) from Fondazione Italiana Sclerosi Multipla; and financed or co-financed with the ‘5 per mille’ public funding. Valsasina, Marchesi, Sona, Tessadori, and Yamin have declared no conflicts of interest.

Acknowledgements:

This study was partially supported by a grant from Fondazione Italiana Sclerosi Multipla (FISM/2018/R/5) and financed or co-financed with the ‘5 per mille’ public funding.

Citation:
EMJ Neurol. ;9[1]:34-36. Abstract Review No. AR1.
Keywords:
Functional connectivity (FC), machine-learning, multiple sclerosis (MS), network analysis, phenotypes, resting state (RS).

Each article is made available under the terms of the Creative Commons Attribution-Non Commercial 4.0 License.

BACKGROUND AND AIMS

Multiple sclerosis (MS) is a chronic, inflammatory, and neurodegenerative disease of the central nervous system, characterised by non-uniform clinical manifestations. Resting state (RS) functional MRI (fMRI) studies of patients with MS often showed trends towards higher functional connectivity (FC) at the earliest disease stages, followed by a gradual RS FC reduction in progressive MS (PMS).1,2 However, complex patterns of regional increased and decreased RS FC in critical brain networks have been described across MS phenotypes. Moreover, RS FC abnormalities do not have a straightforward relationship with the severity of clinical and cognitive symptoms.1,2 Network-based methods on RS fMRI data are helping to shed light into brain functional reorganisation in MS3,4 and, combined with machine-learning, may produce important clues to support automated classification of patients with MS. Against this background, the aim of this study was to develop advanced machine-learning methods to analyse RS FC data and classify patients with MS according to their disease phenotype.

MATERIALS AND METHODS

RS fMRI scans were obtained from 46 right-handed healthy controls (HC) and 113 MS patients (62 relapsing-remitting MS [RRMS] and 51 PMS).

After RS fMRI pre-processing, RS FC matrices were created by means of pair-wise covariance between RS fMRI time series extracted from the automatic anatomical labelling atlas. RS FC matrices were then clustered into groups with dominant-set clustering exploiting the geodesic metric. Each cluster represents a group of subjects with some similarity in their FC. A reference connectome (geodesic mean) was determined for each cluster, and each RS FC matrix was then represented by a set of geodesic distances from cluster representatives. This representation was used to classify subjects (HC, RRMS, and PMS) using a linear support vector machine, executed 100 times with a 5-fold cross-validation scheme. Finally, a sensitivity analysis on the trained classifier was used to identify clusters and connections more relevant for classification.

RESULTS

The described machine-learning tool was able to classify patients with RRMS from HCs with an accuracy of 72.5%, patients with PMS from HCs with an accuracy of 85.2%, and patients with PMS from patients with RRMS with an accuracy of 76.0% (Figure 1). The sensitivity analysis on trained support vector machines found that increased RS FC in the basal ganglia subnetwork (peculiarly involving the bilateral thalami) and abnormal RS FC in the frontal subnetwork contributed to an accurate classification of both patients with RRMS and PMS from HCs (Figure 1). Moreover, decreased RS FC in occipito-temporal subnetworks contributed to differentiate patients with RRMS from HCs, while decreased RS FC in the parietal sub-network (especially involving the precunei and posterior cingulate cortices) contributed to differentiate patients with PMS from patients with RRMS.

Figure 1 Results from machine-learning tool to classify patients with relapsing-remitting multiple sclerosis and progressive multiple sclerosis from healthy controls

Figure 1: Results from machine-learning tool to classify patients with relapsing-remitting multiple sclerosis and progressive multiple sclerosis from healthy controls.
A) Boxplot representing the average cross-fold validation accuracy for the following classification tasks: HCs versus RRMS (blue), HCs versus PMS (orange), and patients with RRMS versus patients with PMS (grey). B,C) Illustrative example of RS FC differences within the main sub-networks of the brain (parcellated according to cortical lobe) contributing to the classification task: B) RS FC differences between HCs and patients with RRMS. C) RS FC differences between HCs and patients with PMS. RS FC decrease in patients versus HCs is colour-coded in blue, while RS FC increase is colour-coded in orange-red.
FC: functional connectivity; HC: healthy control; PMS: progressing multiple sclerosis; RRMS: relapsing-remitting multiple sclerosis; RS: resting state; vs: versus.

CONCLUSION

A combination of different machine-learning principles allowed to classify patients with MS with different clinical phenotypes from HCs with a good accuracy. Increased RS FC in the basal ganglia sub-network contributed to an accurate classification of both patients with RRMS and PMS from HCs, suggesting that altered thalamic and frontal RS FC may be a hallmark of MS disease, occurring in all MS phenotypes. Conversely, the peculiar involvement of occipito-temporal subnetworks in patients with RRMS might be secondary to damage of associative sensory regions, while the peculiar involvement of the posterior cingulate cortex and precuneus in PMS suggests a spreading of damage to high-order, associative regions, leading to impaired network integration.

References
Rocca MA et al. Functional network connectivity abnormalities in multiple sclerosis: correlations with disability and cognitive impairment. Mult Scler. 2018;24(4):459-71. Sbardella E et al. Resting-state fMRI in MS: general concepts and brief overview of its application. Biomed Res Int. 2015;2015:212693. Eijlers AJ et al. Increased default-mode network centrality in cognitively impaired multiple sclerosis patients. Neurology 2017;88(10):952-60.

Rate this content's potential impact on patient outcomes

Average rating / 5. Vote count:

No votes so far! Be the first to rate this content.

Thank you!

Please share some more information on the rating you have given